UnconstrainedOptimization III
UnconstrainedOptimization III
Unconstrained Optimization
Shirish Shevade
Computer Science and Automation
Indian Institute of Science
Bangalore 560 012, India.
min f (x)
s.t. x ∈ Rn
Definition
x∗ ∈ Rn is said to be a local minimum of f if there is a δ > 0
such that f (x∗ ) ≤ f (x) ∀ x ∈ B(x∗ , δ).
15
f(x1,x2)
10
0
3
2
3
1 2
0 1
−1 0
−1
−2 −2
−3 −3
x2 x1
f(x1,x2)
−0.5
2
1 2
1
0
0
−1
−1
−2 −2
x2 x1
2.5
1.5
1 .1 0.1
−0 −0.2 0.2
−0.1
−0
0.1
x2
.3 0.3
0.5
.4
0.
−0.2
0.2
−0
4
−0
−0
.3 3
0.
0.1
−0.5
.1
.2
0.2
−0
−1
−0.1 0.1
−1.5
−2
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2
x1
Proof.
Let x∗ be a local minimum of f and g(x∗ ) 6= 0.
Choose d = −g(x∗ ).
Therefore, g(x∗ ) = 0.
exp(−x12 − x22 )(1 − 2x12 )
g(x) = .
exp(−x12 − x22 )(−2x1 x2 )
g(x) = 0 at ( √12 , 0)T and (− √12 , 0)T .
f(x1,x2)
−0.5
2
1 2
1
0
0
−1
−1
−2 −2
x2 x1
exp(−x12 − x22 )(1 − 2x12 )
g(x) = .
exp(−x12 − x22 )(−2x1 x2 )
g(x) = 0 at ( √12 , 0)T and (− √12 , 0)T .
If g(x∗ ) = 0, then x∗ is a stationary point.
Need higher order derivatives to confirm that a stationary
point is a local minimum
Proof.
Since H is continuous and positive definite near x∗ , ∃ δ > 0
such that H(x) is positive definite for all x ∈ B(x∗ , δ).
Choose some x ∈ B(x∗ , δ). Using second order truncated Taylor
series,
1
f (x) = f (x∗ ) + g(x∗ )T (x − x∗ ) + (x − x∗ )T H(x̄)(x − x∗ )
2
where x̄ ∈ LS(x, x∗ ).
exp(−x12 − x22 )(1 − 2x12 )
g(x) = .
exp(−x12 − x22 )(−2x1 x2 )
g(x) = 0 at x∗1 T = ( √12 , 0)T and x∗2 T = (− √12 , 0)T .
√ 1
2 2 exp(− ) √ 0
H(x∗2 ) = 2 is positive
0 2 exp(− 12 )
definite ⇒ x∗2 is a strict local minimum
√
−2 2 exp(− 12 )
∗ √ 0
H(x1 ) = is negative
0 − 2 exp(− 21 )
definite ⇒ x∗1 is a strict local maximum
2x1 + exp(x1 + x2 )
g(x) = .
exp(x1 + x2 )
Need an iterative method to solve g(x) = 0 .
kg(xk )k ≤
f (xk ) − f (xk+1 )
≤
|f (xk )|
kxk+1 − x∗ k
lim = β, β < ∞
k→∞ kxk − x∗ kp
kxk+1 − x∗ k kxk+1 − x∗ k
lim = 0 and lim =∞
k→∞ kxk − x∗ k k→∞ kxk − x∗ k2